System Requirements
Before you install OnVeris, check that your machines meet the requirements below. The Gateway needs only modest resources. What a Node needs depends on the AI models you intend to run, and a GPU makes a large difference to speed.
Supported Platforms
The Gateway, the Node and OnVeris Chat each run on the platforms listed below.
Gateway And Node
| Platform | Architecture |
|---|---|
| Windows | x86-64, ARM64 |
| Linux | x86-64, ARM64 |
| macOS | Apple silicon (ARM64) only |
Note
macOS on Intel is not supported.
OnVeris Chat
| Platform | Requirement |
|---|---|
| Web browser | Any current browser — served by the Gateway, nothing to install |
| Windows | Windows 10 (version 2004) or later, x86-64 |
| macOS | Apple silicon |
| Linux | x86-64 |
| iPhone and iPad | iOS 15 or later |
| Android | Android 7.0 or later |
See Getting the App.
OnVerisGateway
The Gateway is a lightweight server process with modest resource requirements.
| Minimum | |
|---|---|
| CPU | 2-core, 64-bit |
| RAM | 2 GB |
| Disk | 10 GB, plus room for conversations, attachments and Knowledge Bases |
| Network | A fixed IP address or hostname reachable by users and Node machines |
OnVerisNode
Node requirements depend on the AI models you intend to run. Model files range from a few gigabytes to over 40 GB, and loading them needs either enough system RAM or a GPU with enough VRAM.
| Minimum | Recommended | |
|---|---|---|
| CPU | 4-core, 64-bit | 8+ cores |
| RAM | 8 GB | 32 GB or more, depending on model size |
| GPU | Not required | Dedicated GPU with 8 GB+ VRAM, strongly recommended |
| Disk | 20 GB + model storage | SSD with 100 GB+ available |
GPUs
Running models on a GPU is many times faster than on the CPU. For production use, or any model larger than about 7 billion parameters, a GPU is strongly recommended.
| GPU | Supported through | Needs |
|---|---|---|
| NVIDIA | CUDA 13, CUDA 12 or Vulkan | The NVIDIA driver |
| AMD | ROCm or Vulkan | The AMD driver. On Linux, ROCm also needs ROCm 7.x installed |
| Intel | Vulkan | The Intel driver |
| Apple silicon | Metal | Nothing extra |
On Windows and Linux ARM64 machines, models run on the CPU only.
See Hardware and GPU for how much memory models need.
Network
The Gateway and Nodes use the following connections. Make sure your firewalls allow them.
| Connection | Port | Notes |
|---|---|---|
| Users and applications → Gateway | 443 (HTTPS) | Changeable in Settings. If another program already has 443, the Gateway uses 8025 instead |
| Gateway → Node | 8022 (encrypted WebSocket) | Open on the Node machine's firewall |
| Gateway ↔ Node discovery | 5353 (UDP, mDNS) | Only needed for Auto-discover nearby nodes. See Managing Nodes |
| Gateway → internet | 443 | Licence activation and checks, and the Connector marketplace. Not needed for a licence installed without internet access |
| Node → internet | 443 | Downloading backends and models. Without it, copy model files in by hand |
The installer can open the incoming ports on the Gateway and Node machines for you. See Firewall.